How Energy-Aware Software Can Drive Greater Energy Efficiency

This audio was created using Microsoft Azure Speech Services

We’ve made great strides in increasing energy efficiency through technologies such as virtualization, which dramatically reduces the number of servers companies need. But in many cases, the energy efficiency gains may not be what they could be.

What’s been largely missing is software that’s designed with energy efficiency in mind. The Green Grid is taking steps to address the issue, having launched a Software Task Force whose mission is to raise awareness of the issue, identify and overcome inhibitors and generally educate the IT community about the need to pay more attention to energy efficiency as we develop software.

Mark Aggar, a Green Grid Board Member and Sr. Director of Sustainability at Microsoft, gave a presentation on the topic at the recent Green Grid Forum 2013 event. He pointed out that while virtualization has helped to some degree, applications aren’t being designed or configured to take advantage of fungible, low-cost infrastructure.  We’re also replacing mostly idle physical machines with mostly idle virtual machines.

It’s a real problem because when servers are under-utilized they don’t scale useful work performed to energy consumed, meaning they chew up power at a rate that’s not commensurate with the work they’re performing. An idle server still consumes about 32% of its peak power load while a server at 10% load consumes about 50% of its peak power. At 70% load, however, the server consumes roughly 79% of peak power. That means if you can increase utilization from 10% to 70%, you get 7 times the output for only 58% more power.

If we had software that was smarter about how it used energy, we could increase the efficiency of servers while getting more output from them. In the end, we’d need less infrastructure.

Here’s one example that Aggar used in his presentation. Suppose you’ve got 100 servers dedicated to stock trades, and another 100 dedicated to video streaming, with each only occasionally reaching 50% utilization. If you instead go with 110 servers running both video and stock trading, you’ll need 45% fewer servers – and you’ll still have lots of idle capacity, roughly 50%.  Now if you add smart software that can identify when a server has spare capacity and dynamically shift workloads to it, you can fill that capacity.

There are lots of other ways developers can create more energy efficient software, such as taking steps to help systems stay idle, measure workload power efficiency, scale resource use intelligently, enabling or configuring power management (not just turning it off) and more.

Companies like Schneider Electric are interested in this area because we see an opportunity to drive further energy efficiencies.  Our data center infrastructure management offerings already have the ability to reach into IT gear and monitor how the software is running. But if microprocessors and software start reaching out to us to report on various elements, it will help make our DCIM offerings that much more valuable.

That kind of technology is a few years away, but believe me – we’re working on it.